Image Classification and Text Identification in Inspecting Military Aircrafts Logos: Application of Convolutional Neural Network

Object detection and inspection using images or videos have been receiving increased attention in many applications such as traffic control, brand monitoring, trademark compliance, and product authentication. A particular application that is currently a topic of interest is aircraft logo detection, which aims at automating the visual inspection carried out manually by aircraft engineers. Aircraft logos should meet a large set of requirements that include geometric constraints on the logo elements and patterns, and constraints on the position and orientation with respect to specific references. This work considers the design of a high accuracy convolutional neural network to detect and classify aircraft logos as either adequate or inadequate based on specified criteria. The performance of the developed network is compared to a number of classical machine learning algorithms to demonstrate its effectiveness. Adequate logos are then processed further by extracting them from a frame using robust features extraction algorithm and determining their orientation angle with respect to the horizontal reference axis. Afterward, a text detection technique using a character region awareness for text detection algorithm implemented on a pre-trained network is carried out, along with optical character recognition tool to detect and extract the text from the logos for further processing in other applications. The developed network is tested on actual aircraft logos, captured from the field, where satisfactory results are obtained.